Numbers and US

Story that numbers tell us

YouTube Recommendation Engine

leave a comment »

While watching a YouTube video I was pleasantly surprised to see a video recommendation. I was watching a Hindi song, and got recommendation of an interview of P!nk on a topic that was highly relevant to the theme of the song. If you have seen the movie Abhiman, you are certainly going to be impressed with the recommendation!

recommendation Engine

Have YouTube statistical modelers  made their recommendation engine advance enough to recommend us videos based on our mood/sentiment?  I doubt that.

To understand, lets just think of the data YouTube collected from my activities. I watched couple of P!nk video. I am not sure whether I listened to any Hindi songs in the last couple of weeks, but that’s not of enough relevance here.  Now the possible hypothesis could be:

a:) YouTube knows I watch P!nk videos along with many more videos. Quite possible that randomly, just by sheer chance, it recommended me to watch one of the P1nk video.   Relevance of sentiment was just a fluke.  Well, this is always a possibility, and in fact, number of times chance and randomness are answers to so many puzzles we bump our head to. But, I am positive, recommendation engine is smarter than this.

b:) It’s quite possible, YouTube might have bucketed all the videos in their database based on ‘sentiments’. The video from Abhiman might have been bucketed under the same ‘sentiment’ as the interview of P1nk. Hence, the moment I watch the Abhiman video, YouTube recommended me a video  with the same sentiment, and of someone I watch.  I would guess YouTube might have started using this approach for number of recommendation, but I have some reservation around how well they might be using the approach. It’s difficult as you have to bucket a video in millions of group across multiple dimension. Sometime user generated ‘sentiment group’ is the answer, but getting as much data as you really want is a challenge.

c:) The third possibility could be the  approach that was, and it still is, the core of most of recommendation engines. There might be someone who would have watched the Abhiman video, and the same guy might have watched the P1nk video as well. Millions would have watched Abhiman, and most of them won’t have watched P!nk video subsequently. So, the recommendation rule tagged me along with a guy who watched P1nk video, as I too like P!nk video.

If you think about it, the real business rule could be a combination of any of the three hypothesis, including the one based on randomness. But, in any case, it was nice to  get the recommendation as I really ended up watching it.


Written by SK

November 6, 2013 at 7:04 am

Posted in Uncategorized

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

Humor, Sarcasm from & on Silicon Valley

Let's have a laugh together

Product Thinking

Peeling the layers of products that delight is the best place for your personal blog or business site.

%d bloggers like this: